1. Field of the Invention
The present invention relates to a packet switching system and in particular to a pipelined scheduling method and scheduler implemented in the packet switching system.
2. Description of the Prior Art
With an explosion in the use of the Internet, there is a growing demand for using the Internet as an infrastructure of communications. In order that the Internet acts as such a communication infrastructure, it is necessary for a router as a server node to enhance the speed of data transfer and its function. Existing high-speed routers employ IP address searching achieved with hardware and high-speed data transfer processing with a self-routing high-speed switch fabric.
To meet such a growing demand for high-speed switching, there has been widely used a Virtual Output Queuing (VOQ) crosspoint switch having N input ports and N output ports, where each input port has N logical queues each corresponding to the N output ports. There have been proposed scheduling methods for such a crosspoint switch.
A two-dimensional round-robin scheduling mechanism has been disclosed in U.S. Pat. No. 5,299,190. This mechanism uses a request matrix with each row representing an input and each column representing an output. A bit in a given row and column of the matrix thus represents a request from a corresponding input port for connection to a corresponding output port. Diagonal service patterns are used to overlay the request matrix to determine which requests are to be serviced. A sequence of diagonal service patterns for each of K time slots is used to provide guaranteed service with fairness.
A similar scheduling method in a data packet router has been disclosed in U.S. Pat. No. 5,734,649. In the data packet router, a matrix of crosspoint switch elements connects data sources to selected destinations during each of a succession of intervals. Allocation of switch elements to desired connections is accomplished by a process which provides a data array having a number of data elements corresponding to the number of switch elements. During each interval, a source is assigned to each of the data elements in accordance with a first current pseudo-random shuffle pattern and a respective destination is assigned to each of the data elements in accordance with a second current pseudo-random shuffle pattern. A new set of shuffle patterns is generated during each interval. A testing of successively progressing diagonal grouping of regions across the array of the sources and destinations is performed to search for a match not previously allocated and each match is allocated to the switch element corresponding to the respective data element.
However, these conventional scheduling methods have a disadvantage such that the volume of data processing for each time slot goes up with the square of the number of ports. Accordingly, it becomes difficult to enhance the speed of scheduling as the number of input/output ports increases.
As a promising scheduling protocol to overcome the above disadvantage, a round-robin greedy scheduling (RRGS) algorithm has been proposed by the present Applicant in Japanese Patent Application No. 11-172584 (Unexamined Publication No. P2000-174817) . The RRGS algorithm can be implemented in an N×N packet switch, wherein N scheduling modules S1–SN are provided for N inputs, respectively. Each of the scheduling modules S1–SN performs scheduling for a predetermined future time slot, and transfers reserved output port information to the adjacent scheduling module. In this way, output port reservation at the predetermined future time slot can be completed during N time slots prior to the predetermined future time slot. Such scheduling is performed at each time slot by pipeline processing to achieve N×N scheduling for future time slots, resulting in high-speed packet forwarding.
A framed RRGS algorithm has been proposed by the present Applicant in Japanese Patent Application No. 2000-55103 (Unexamined Publication No. 2001-7822). In the framed RRGS, a sequence of frames is set, each of which consists of a plurality of time slots. Input packets are scheduled in a current frame so that they are forwarded to appropriate ones of the output ports in a next frame following the current frame. More specifically, the scheduling in each frame is performed by simultaneously starting scheduling decision processes of the N input port scheduling modules at the beginning of the frame, simultaneously performing the scheduling decision processes using a pipelined approach in the frame, and simultaneously completing the scheduling decision processes at the end of the frame.
Although the above-described RRGS and framed RRGS algorithms can provide high-speed and high-efficient data forwarding, the amount of output-port arbitration processing increases as the number of input and output ports increases.